Read the cluster assignments, parameters, and summary for a spatial method from a Seurat object's stored tool result. If detailed results were not stored, cluster assignments are read from the corresponding metadata column. BANKSY retains a lightweight index for custom result columns and exact sample membership. Sample retrieval requires this index or stored per-sample results; sample identity is never inferred from cluster labels.
Results describe cells present in object, in its current cell order, rather
than the original fit's full cell set. BANKSY domain counts and PRECAST cell,
domain, and sample counts are recomputed from the returned assignments.
Unfitted PRECAST cells are excluded. Parameters and the native bundles in
object@tools still describe the original fit; other methods' stored summaries
are omitted when their cell set changes. PRECAST can be selected by its tool
name or by "PRECAST" when that method identifies a unique stored bundle.
New BANKSY fits and detailed SmoothClust, CHOIR, RareQ, and PRECAST results
retain a private metadata identity column, named by the result index.
Preserve this column to retrieve results after subset() or RenameCells();
names are mapped by saved identities, never by order or sample prefixes.
Missing, duplicate, or unknown identities cause an informative error.
SmoothClust and CHOIR's explicit cell columns also reflect the current IDs.
Legacy saved assignments support exact-ID subsets. Contradictory metadata
causes an error, but renames cannot always be detected without provenance;
rerun the method to establish identity tracking. Legacy PRECAST and
metadata-only results use their current metadata assignments.
Storage-disabled methods other than BANKSY retain their existing storage behavior. Custom metadata columns require a stored locator; the accessor does not guess a column from its values or prefix.